mHealth Opportunities for Physician-Scientists to ...165k mHealth apps, while
Participatory mHealth - microsoft.com€¦ · – User engagement/experience: motivate sustainable...
Transcript of Participatory mHealth - microsoft.com€¦ · – User engagement/experience: motivate sustainable...
Smartphones are proximate, pervasive, programmable, personal; offer data/interaction in real time, real place, real context
Participatory mHealth
Deborah EstrinUCLA Center for Embedded Networked Sensing (CENS)
and OpenmHealth.org (~fall 2011). contact: [email protected]
in collaboration with faculty, students, staff at CENS, UCLA, UCSF...
Stovepipe Architecture
Mobile platformsiPhone/Android /Feature Phones
Patient /Caregivers Patient /Caregivers
Open mHealth Architecture
Mobile platformsiPhone/Android /Feature Phones
Re-usable health data and knowledge services
Standardized personal data vaults and health specific data exchange protocols
Analysis /visualization/
feedback
Processing
Storage
Data transport
Data capture
AP
PL
IC
AT
IO
NS
1Thursday, July 14, 2011
experience sampling streams
personal data repository
event detection
processing
aggregate measures, trends, patterns
our self report
visualization
our actions
Photo: Marshall Astor, WWW
context and activity traces
mHealth
Use mobile devices to enhance health and wellness by extending health interventions and research beyond the reach of traditional clinical care.
2Thursday, July 14, 2011
motivation3 behaviors (diet, exercise, smoking) cause 1/3rd of US deaths
50% Americans have 1 or more chronic diseases
age of onset getting younger
chronic disease prevention/management/research happens in the context of daily life, outside of clinical setting
approachsupport individuals, communities, clinicians to continuously improve patient-centered, personalized, health and healthcare.
mobile devices offer proximity, pervasiveness, programmability, personalization, affordability.
Why participatory mHealth ?
3Thursday, July 14, 2011
Dashboard access
Internet
Data collec0on
Data analysis, privacy preserving storage
Ramanathan, Selsky, et al
built to explore participatory mHealth
4Thursday, July 14, 2011
Phonetop Bu:ons
Internet
GPS ACCELEROMETER
Ac0graphy
Loca0on and 0me triggered
surveys
Automated capture
Data collec0on
Ramanathan, Selsky, et al
5Thursday, July 14, 2011
Participatory design: functionality shaped by focus groups, interviews
6Ramanathan et al
AndWellnessApplicationServer
<survey> <title>Alcohol</title> <prompt> <text>How many drinks did you have today?</text> </prompt></survey>
Survey Authoring
Image Capture
Stress, Eating Buttons
GPS ACCELEROMETER
Smart Classifier
Actigraphy
Survey Input
Trigger Authoringtrigger logic as function of
location, time, activity, prompt responses
>100 (somewhat) diverse participants: young moms, young men living with HIV, immigrant women, breast cancer survivors, and recruited UCLA student testers
6Thursday, July 14, 2011
Focus groups summary (n=72)
Population Primary application features discussed
Young Momsn=23
Engaging participation without increasing user burden•Customization of reminder times/locations for convenience•Image capture of food to increase accountability•Set, manage, and monitor progress towards a goal•Light-weight data capture•Primary interaction to take place on the phone
Immigrant Womenn=20
Mechanisms to help people achieve a goal•Set, manage, and monitor progress towards a goal•Helpful tips and problem solving (suggested by phone)
People Living with HIV
n=29
1) Privacy and security of data•Password protection on phone a must•Nondescript text to hide the intent of sensitive questions•Location tracking is controversial, granular control a must•Data anonymization for sharing with counselor, coach, medical provider
2) Customizatin of Reminders • Medication adherence reminders, especially using location• Safe sex reminders
Ramanathan, Swendeman, Dawson, Estrin, Rotheram-Borus7Thursday, July 14, 2011
Notable feature requests
• Images: Moms LOVED this feature for food, SA women did not.
• Triggers: Control of timing important to all--need trigger authoring and personalization
• Buttons: Most moms willing to answer at least briefly ‘in the moment’, while SA women almost all wanted to answer only at the end of the day.
• Feedback: Very few interested in seeing simple quantifications of their responses. Helpful tips and motivational messaging most popular. SA explicitly preferred *against themselves* vs competitive feedback with group.
• Server vs Phone: Very few willing/interested to access server. Most wanted interaction solely on phone.
Ramanathan, Swendeman, et al
8Thursday, July 14, 2011
End-‐User Dashboards
Correla0ons in 0me and space
Internet
Work in progress for future release: time delayed correlations and correlations across behaviors
Ramanathan, Selsky, et al
Ac0graphy over space Ac0graphy over 0me
9Thursday, July 14, 2011
Feedback
Configure
Chart Summaries
Table Summaries
Iyer, Ramanathan, et al
* To Be Released August, 2011
10Thursday, July 14, 2011
ubifit garden using on-body sensing, activity inference, and a personal, mobile display to encourage regular and varied physical activity
2 dub, u of washington
1 intel research seattle
had the garden
participants who…
did NOT have the garden
consolvo, mcdonald, landay, chi 09 consolvo et al, ubicomp 08 consolvo et al, chi 08 choudhury et al, ieee pervasive mag ‘08 froehlich et al, mobisys ‘07 consolvo, paulos, smith, mobile persuasion ’07
Of course...“feedback” should look more like this and be tailored to individual participants-- Ubifit (UW, Intel)
11Thursday, July 14, 2011
Researcher Dashboards
Survey Crea0on
Internet
Par0cipa0on Popula0on Sta0s0cs
* in progress for release, Dec, 2011
05
1015
2025
0.0 0.5 1.0 1.5 2.0 2.5 3.0Survey response rate
Num
ber o
f day
s
Ramanathan, Selsky, et al
12Thursday, July 14, 2011
Admin Dashboards
Internet
User interac0on with phone
Remaining ba:ery on phone
Memory usage on phone
Filter by loca0on, 0me, user
Wu, Falaki, Estrin, Ramanathan13Thursday, July 14, 2011
andwellness.org
Java 6 + Tomcat 6 + MySQL 5.1
Internet
AndWellness system implementation
14
Server
JSON/HTTPS
JSON/HTTPS
Smartphone
Web Browser Clients
AndWellness
andwellness.org
Java 6 + Tomcat 7 + MySQL 5.1
RApache for datavis
Ramanathan, Selsky, et al
What lies behind exposed API calls could be written in any programming language.
Server-side defined by HTTP APIs: GET and POST using JSON, standard HTTP data packaging
14Thursday, July 14, 2011
• Experience sampling, light-weight data capture (50%)
• Visualization/data presentation for end-user and researcher (15%)
• Smart triggers based on user configurable location, time, activity (25%)
• Background services for actigraphy, location tracing, system analytics (50%)
• Battery-preserving background services (50%)
• Open mHealth community building (15%)
• Participatory privacy policies and mechanisms (15%)
15
Status of AndWellness features to date(with guesstimate on how far along we are)
Ramanathan,, et al
15Thursday, July 14, 2011
Key research challenges
• Health sciences community:
– Establish validity and reliability of mHealth instruments
– Derive efficacy evidence base from rich usage, system analytics
– Behavior change: defining, implementing, and adapting interventions that support sustained and beneficial change across populations
• Technical community:
– Infovis: analysis, presentation, visualization, for self, clinician, researcher
– Resource management, efficiency (enable full-day phone operation with background activity and data capture)
– User modeling (eg community models (Choudhury) for activity classification, context, triggers
– User engagement/experience: motivate sustainable user participation with game mechanics, adaptive interfaces
– Selective sharing, usable privacy tools
– Open systems16
Ramanathan, et al16Thursday, July 14, 2011
Open mHealth System: functional components
EHR/PHR, social media
Data Exchange standards
Import and
Export
Dashboards,Presentation
Mobile Data Collection
Analysis Services
Configuration and
management
Personal dashboardsClinical
Community/social group dashboards
Reminders, Triggers
Prompted self-
reports
Automated measure-
ments (actigraphy, offboard de-
vices)
Feedback, social me-dia, goal setting
DataStore
Data Management
SecurePersonal
Data Vault
Core ID, Security Services
Data alignment
Time series summarization
Summary statistics
correlations
Trend and anomaly
identificationAggregateanalysis
InfoVis
External data layers
Game and guided-in-tervention
stats
Phone us-age patterns (communi-
cations, calendars,
alarms)Current
Stacklets
Activity classification
InfoVisextract and present relevant trends, patterns, anomalies, correlations across
diverse data streams and to diverse audiences
Needs: pre-processing, feature extraction, integration with machine learning libraries and statistical analysis tools, incorporation of external datasets, geo-spatial analyses, informative and configurable presentation
17Thursday, July 14, 2011
•Battery and resource monitor continuously guides applications to consume enough power to meet the deadline; trading off fidelity/resolution
Adaptive battery management for background applications
Falaki, et al
Energy Source
Interaction Metrics
Phone Resources
•Usage and context matter for battery management, e.g., 15% left battery at 10pm is not the same as 15% at 10am.
SystemSens
{GPS ACCELEROMETER
Andwellness
Energy Budget
18Thursday, July 14, 2011
19
Objectives• energy-efficiently sense semantic locations on battery-limited mobile devices• automatically learn and recognize semantic places and paths closer to user’s
interpretation of location• motivate user feedback to bridge between machine-learned and human-defined places
Selectively leverage GPS/Wi-Fi/Accelerometer when each is informative/efficient• people spend approximately 89% indoors, 5% in a vehicle, and 6% at outdoors on average
Signal patterns of surrounding RF beacons are
noisy but useful for detecting semantic places
Sensor traces from a single day following a normal routine
GPS position fixes are inaccurate most of the time but
informative when traveling paths connecting places
Many power saving opportunities exists
when the device is immobile
Semantic places and path
19Thursday, July 14, 2011
User modeling (sic) Using community similarity networksto handle population diversity (T. Choudhury, Cornell)
A
B
C
D
Data
Labels&
SimilaritySensitiveTraining
PhysicalSimilarityNetwork
LifestyleSimilarityNetwork
Sensor DataSimilarityNetwork
Similarity NetworkCo-Training
Community AModel
Community BModel
Community CModel
Community DModel
20Thursday, July 14, 2011
User engagement:informational incentives, feedback, game mechanics
Informational incentives, e.g., analytics about actions, encourage participation initially--See Consolvo, Choudhury, Mynatt
Micro-payments/rewards promoted even (sustained) participation in community data gathering--might also apply for participatory mHealth Reddy, et al 2009:• Micro-payments based on competition worked best for short bursty data collections• Very low baseline micro-payments discouraged individuals
Future directions: game mechanics, social media tie-ins, goal setting and monitoring tools, adaptive over time for sustainability, configurable
Mobile Ambient Wellbeing Display (T. Choudhury, Cornell)
21Thursday, July 14, 2011
Mobile App Personal Data Vault
Third Party Services
- Data Capture / Upload(Prompted, Automated)
- Reminders
- Feedback, Incentives
Well defined interfaces allowmobile functions to be mixed,matched, and shared
- User Identity and Authentication
- Long-term Data Management
Patient defined selective sharing with Open mHealth Server function
- Analytics for PersonalData Streams
- Interface to Clinical CarePlan, Personnel
- Integration with EHR/PHRs
- Cross Patient Aggregation
Well defined interfaces allowanalytics functions to be mixed,matched, shared, compared
Personal Data Vault (PDV): allow participants to retain control over their raw data by
decoupling capture and sharing
vault + filters = granular, assisted control over what/when you send to whom, what data says about you, whether you reveal who you are or share anonymously, ... M. Mun, et al, CONEXT 2010
22Thursday, July 14, 2011
But...can we make selective sharing usable?
PDV
1. Upload raw data
Content-Service Providers
3. Request data
4. Filter data to send
5. Send the filtered data
6. Check easy-to-read reports about continuous data sharing
Trace-audit
Step 6 and 7 can be repeated
7. Reconfigure ACLs
Rule Recommender
2. Configure privacy policies/ACLs
2. Configure privacy policies/ACLs
Granular ACLs
M. Mun, et al, CONEXT 2010
bounds, precision, frequency
across time, space,
Significant Location
Identification
Routine Place Selection
Spatial Bound Creation
Rule Recommender
Granular ACLs:
23Thursday, July 14, 2011
Importance of an open platform: avoid silos, promote innovation and transparency
Bootstrap rapid cycle of learning, sharing, deployment-~80 % (guesstimate) system components reusable-Largest missing pieces: authoring, analysis-visualization, feedback
Facilitate research in methodology, treatment-Systems gather usage data automatically for evaluation, iterative improvement-Encourage modularity and sharing in methodologies, practice
Development in the context of real applications and use-Collaborative/participatory design process with continual feedback from users-Diverse targeted pilots inform generalization, adaptation, expansion.
Explore balancing of privacy protection and data sharing-Variety of privacy/sharing policies-STransparency of research and data processes for participants
24Thursday, July 14, 2011
Open mHealth System: functional components
EHR/PHR, social media
Data Exchange standards
Import and
Export
Dashboards,Presentation
Mobile Data Collection
Analysis Services
Configuration and
management
Personal dashboardsClinical
Community/social group dashboards
Reminders, Triggers
Prompted self-
reports
Automated measure-
ments (actigraphy, offboard de-
vices)
Feedback, social me-dia, goal setting
DataStore
Data Management
SecurePersonal
Data Vault
Core ID, Security Services
Data alignment
Time series summarization
Summary statistics
correlations
Trend and anomaly
identificationAggregateanalysis
InfoVis
External data layers
Game and guided-in-tervention
stats
Phone us-age patterns (communi-
cations, calendars,
alarms)Current
Stacklets
Activity classification
25Thursday, July 14, 2011
24
Stovepipe Architecture
Mobile platformsiPhone/Android /Feature Phones
Patient /Caregivers Patient /Caregivers
Open mHealth Architecture
Mobile platformsiPhone/Android /Feature Phones
Re-usable health data and knowledge services
Standardized personal data vaults and health specific data exchange protocols
Analysis /visualization/
feedback
Processing
Storage
Data transport
Data capture
AP
PL
IC
AT
IO
NS
“Approximately 25 years ago, government and industry invested in expanded access at a crucial time in the Internet’s development. The resulting networks and ubiquity of access provided fertile ground for technologies, ideas, institutions, markets, and cultures to innovate. The payoff from this investment created a commercially viable and largely self-governing ecosystem for innovation.
The same can be done for global health. Government, commercial, and nongovernmental entities involved in health IT and innovation should cooperate to define and instantiate architecture, governance, and business models and to steer initial mHealth investments into open architecture.”
- D Estrin, I Sim. Open mHealth Architecture: An Engine for Health Care Innovation. Science Magazine, Nov, 2010.
Closing remarks (aka parting shots)
Open mHealth initiative: http://openmhealth.org
Summer reading recommendations:
The filter bubble, Eli Pariser
Everything is Obvious*, *once you know the answer, Duncan Watts
26Thursday, July 14, 2011
Acknowledgments: Collaborators and Sponsors
Collaborators
Technology faculty, PIs: Deborah Estrin, Mark Hansen, Nithya Ramanathan
Application/domain expert faculty/PIs (Health science): Robert Bilder, Jacqueline Casillas, Scott Comulada, Patricia Ganz, Mary Jane Rotheram-Borus, Ida Sim (UCSF), Fred Sabb, Dallas Swendeman, Michael Swiernik
Students, staff: Staff: Betta Dawson, John Jenkins, Mo Monibi, Joshua SelskyGraduate students: Faisal Alquaddoomi, Hossein Falaki, Brent Flagstaff, John Hicks, Jinha Khang, Donnie Kim, Min Mun, Katie Shilton
Sponsors and Partners/Collaborators
UCLA centers: CENS, Global center for families and children, Health Sciences, JCCC Federal funding: NSF STC and NETS-FIND Program, NIH
Corporate funding: Google, Intel, MSR, Nokia, T-Mobile Foundations/NGOs: The California Endowment, RWJF, CHCF, CRA
27Thursday, July 14, 2011